This thesis aims at developing an adaptive and nonlinear model predictive control Simulink scheme and interfacing it with the popular PX4 drone system. PX4 is one of the most used drone \ac{RTOS} in the context of research, it has many safety and sensor management features, it is open source, and has an extensive and active community of developers making it an excellent platform for \ac{UAVs} control development. The advantages of interfacing it with Matlab/Simulink running on a companion computer are mainly twofold. The first is simplicity: the Simulink block scheme language is easy to use for complex control schemes, also supported by a great collection of libraries and by the baked-in management of PX4 of sensor data that can directly be used as feedback for the controls without additional estimators. The second is the possibility of moving the computational complexity away from the onboard embedded platform to a much more powerful ground station PC. \ac{NMPC} is an excellent example as it makes use of both, there are many implementations available that require only some setup and the model of the plant, it gives great control performance but is computationally expensive and therefore not always usable directly of low-end embedded hardware without some optimizations, which would require a competent and experienced user. Since model predictive control is susceptible to modeling errors that are especially common when dealing with low-cost drone platforms it is paired with a lightweight adaptive scheme that complements the control action to make up for modeling mismatches. The whole infrastructure is then validated through \ac{SITL} simulations across a variety of tasks and conditions, confirming that the interface between Matlab/Simulink works, the \ac{NMPC} scheme is usable in real-time with good trajectory tracking performance and that adaptive control provides a much greater degree of robustness to the system.

Implementation of adaptive nonlinear model predictive control on a PX4-enabled quad-rotor platform

BONAZZA, MARCO CONCETTO
2022/2023

Abstract

This thesis aims at developing an adaptive and nonlinear model predictive control Simulink scheme and interfacing it with the popular PX4 drone system. PX4 is one of the most used drone \ac{RTOS} in the context of research, it has many safety and sensor management features, it is open source, and has an extensive and active community of developers making it an excellent platform for \ac{UAVs} control development. The advantages of interfacing it with Matlab/Simulink running on a companion computer are mainly twofold. The first is simplicity: the Simulink block scheme language is easy to use for complex control schemes, also supported by a great collection of libraries and by the baked-in management of PX4 of sensor data that can directly be used as feedback for the controls without additional estimators. The second is the possibility of moving the computational complexity away from the onboard embedded platform to a much more powerful ground station PC. \ac{NMPC} is an excellent example as it makes use of both, there are many implementations available that require only some setup and the model of the plant, it gives great control performance but is computationally expensive and therefore not always usable directly of low-end embedded hardware without some optimizations, which would require a competent and experienced user. Since model predictive control is susceptible to modeling errors that are especially common when dealing with low-cost drone platforms it is paired with a lightweight adaptive scheme that complements the control action to make up for modeling mismatches. The whole infrastructure is then validated through \ac{SITL} simulations across a variety of tasks and conditions, confirming that the interface between Matlab/Simulink works, the \ac{NMPC} scheme is usable in real-time with good trajectory tracking performance and that adaptive control provides a much greater degree of robustness to the system.
2022
Implementation of adaptive nonlinear model predictive control on a PX4-enabled quad-rotor platform
Failsafe
UAV
Adaptive
Model predictive
File in questo prodotto:
File Dimensione Formato  
Bonazza_MarcoConcetto.pdf

accesso aperto

Dimensione 4.96 MB
Formato Adobe PDF
4.96 MB Adobe PDF Visualizza/Apri

The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12608/46069