Our world is living a paradigm shift in technology policy, often referred to as the CyberPhysical Revolution or Industry 4.0. Nowadays, CyberPhysical Systems are ubiquitous in modern control engineering, including automobiles, aircraft, building control systems, chemical plants, transportation systems, and so on. The interactions of the physical processes with the machines that control them are becoming increasingly complex, and in a growing number of situations either the model of the system is unavailable, or it is too difficult to describe accurately. Therefore, embedded computers need to "learn" the optimal way to control the systems by the mere observation of data. What seems the best approach to control these complex systems is often by discretizing the different variables, thus transforming the model into a combinatorial problem on a finitestate automaton, which is called an abstraction of the real system. Until now, this approach, often referred to as "abstractionbased control" or "symbolic control", has not been proved useful beyond small academic examples. In this project I aim to show the potential of this approach by implementing a novel datadriven approach based on a probabilistic interpretation of the discretization error. I have developed a toolbox (github.com/davidedlucl/masterthesis) implementing this kind of control with the aim of integrating it in the Dionysos software github.com/dionysosdev). With this software, I succeeded in efficiently solving problems for nonlinear control systems such as a path planning for an autonomous vehicle and a cartpole balancing problem. The longterm objective of this project is to improve the methods implemented in my current software by employing a variable discretization of the state space and to consider complex specifications such as LTL formulas.
Our world is living a paradigm shift in technology policy, often referred to as the CyberPhysical Revolution or Industry 4.0. Nowadays, CyberPhysical Systems are ubiquitous in modern control engineering, including automobiles, aircraft, building control systems, chemical plants, transportation systems, and so on. The interactions of the physical processes with the machines that control them are becoming increasingly complex, and in a growing number of situations either the model of the system is unavailable, or it is too difficult to describe accurately. Therefore, embedded computers need to "learn" the optimal way to control the systems by the mere observation of data. What seems the best approach to control these complex systems is often by discretizing the different variables, thus transforming the model into a combinatorial problem on a finitestate automaton, which is called an abstraction of the real system. Until now, this approach, often referred to as "abstractionbased control" or "symbolic control", has not been proved useful beyond small academic examples. In this project I aim to show the potential of this approach by implementing a novel datadriven approach based on a probabilistic interpretation of the discretization error. I have developed a toolbox (github.com/davidedlucl/masterthesis) implementing this kind of control with the aim of integrating it in the Dionysos software github.com/dionysosdev). With this software, I succeeded in efficiently solving problems for nonlinear control systems such as a path planning for an autonomous vehicle and a cartpole balancing problem. The longterm objective of this project is to improve the methods implemented in my current software by employing a variable discretization of the state space and to consider complex specifications such as LTL formulas.
AbstractionBased DataDriven Control
DE LAZZARI, DAVIDE
2021/2022
Abstract
Our world is living a paradigm shift in technology policy, often referred to as the CyberPhysical Revolution or Industry 4.0. Nowadays, CyberPhysical Systems are ubiquitous in modern control engineering, including automobiles, aircraft, building control systems, chemical plants, transportation systems, and so on. The interactions of the physical processes with the machines that control them are becoming increasingly complex, and in a growing number of situations either the model of the system is unavailable, or it is too difficult to describe accurately. Therefore, embedded computers need to "learn" the optimal way to control the systems by the mere observation of data. What seems the best approach to control these complex systems is often by discretizing the different variables, thus transforming the model into a combinatorial problem on a finitestate automaton, which is called an abstraction of the real system. Until now, this approach, often referred to as "abstractionbased control" or "symbolic control", has not been proved useful beyond small academic examples. In this project I aim to show the potential of this approach by implementing a novel datadriven approach based on a probabilistic interpretation of the discretization error. I have developed a toolbox (github.com/davidedlucl/masterthesis) implementing this kind of control with the aim of integrating it in the Dionysos software github.com/dionysosdev). With this software, I succeeded in efficiently solving problems for nonlinear control systems such as a path planning for an autonomous vehicle and a cartpole balancing problem. The longterm objective of this project is to improve the methods implemented in my current software by employing a variable discretization of the state space and to consider complex specifications such as LTL formulas.File  Dimensione  Formato  

DeLazzari_Davide.pdf
accesso aperto
Dimensione
1.84 MB
Formato
Adobe PDF

1.84 MB  Adobe PDF  Visualizza/Apri 
The text of this website © Università degli studi di Padova. Full Text are published under a nonexclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/33179