The introduction of robotics and automation in the field of industrial production, together with the development of innovative process technolo- gies such as the deposition for additive manufacturing or spraying and automated glueing have brought to undoubted advantages and benefits in the sector, increasing the flexibility of the companies, their efficiency and ability to offer increasingly customized and de-standardized. However, the introduction of innovations in consolidated production contexts that move towards a philosophy of Industry 4.0, leads to problems and criticalities that in the past had not been considered, together with fields of research previously unexplored. What we are focusing on is the solution to the problems of Motion Planning that, compared to the past, must now con- sider contexts and constraints that in traditional machines and processes do not occur. For example, the adaptation of robots for machining and other mechanical operations, while offering great flexibility and ability to be implemented at a significantly lower cost of a machining centre, do not guarantee the dynamic performance of machines designed natively for this purpose. In addition, performances are generally not constant throughout the workspace but depend on the configuration assumed by the robot. Moreover, with regard to innovative technologies, the perfor- mances required in processes such as deposition depend on the dynamics of the deposition itself, to guarantee the quality of the final product. This introduces constraints of nature kinematic and dynamics natures in differ- ential form that need to be incorporated in Path and Trajectory Planning problem-solving methodologies. Here a method is proposed to solve the problems of Motion Planning that implements kinematic and dynamic differential constraints and provide a feasible trajectory for the machine that minimizes the geometric deviation between the proposed trajectory and the one theoretically required from the machine. Afterwards, to evaluate and have a methodology benchmark, a different approach is de- veloped by implementing Reinforcement Learning in a physical simulation environment capable to solve the Motion Planning problem using the v techniques of Artificial Intelligence. Finally, the results of the tests are discussed and the strengths are identified and compared, together with the peculiarities of the two approaches.
L’introduzione della robotica e dell’automazione nel campo della pro- duzione industriale, unitamente allo sviluppo di innovative tecnologie di processo come il deposition per additive manufacturing o spraying e glue- ing automatizzati hanno portato a indubbi vantaggi e benefici nel settore, incrementando la flessibilità delle aziende, la loro efficienza e capacità di offrire prodotti sempre più customizzati e de-standardizzati. Tuttavia, l’introduzione di innovazioni in contesti di produzione consolidati che si spostano verso una filosofia di Industria 4.0, porta a problematiche e criticità che in passato non erano state considerate, unitamente a campi di ricerca precedentemente inesplorati. Ciò su cui ci si concentra, è la soluzione dei problemi di Motion Planning che rispetto al passato devono considerare contesti e vincoli che in macchine e processi tradizionali non si presentano. Ad esempio, l’adattamento di robot per operazioni di machin- ing e di altre lavorazioni meccaniche, mentre offrono grande flessibilità e capacità di essere implementati a un costo sensibilmente inferiore di un centro di lavoro, non garantiscono quelle prestazioni dinamiche di macchine progettate nativamente per questo scopo. Inoltre, le prestazioni non sono in genere costanti in tutto lo spazio di lavoro ma dipendono dalla configurazione assunta dal robot. Anche per quanto riguarda tec- nologie innovative, in processi come il deposition le prestazioni richieste alla macchina dipendono dalla dinamica della deposizione per garantire la qualità del prodotto finale. Questo introduce vincoli di natura cinematica e dinamica in forma differenziale che hanno bisogno di essere incorpo- rati nelle metodologie di soluzione dei problemi di Path e Trajectory Planning. Qui viene proposto un metodo di soluzione dei problemi di Motion Planning che implementi vincoli differenziali cinematici e dinamici e che fornisca una traiettoria fattibile per la macchina che minimizzi la deviazione geometrica tra la traittoria proposta e quella teoricamente richiesta alla macchina. In seguito, per valutare e avere un benchmark della metodologia, un approccio diverso viene sviluppato implementando il Reinforcement Learning in un ambiente di simulazione fisica capace iii di risolvere di risolvere il problema di Motion Planning utilizzando le tecniche di Artificial Intelligence. Infine vengono discussi i risultati dei test ed individuati e confrontati i punti di forza e le peculiarità dei due approcci.
Motion planning under kinetodynamic contraints : numerical and machine learning approaches
BERARDINUCCI, FRANCESCO
2018/2019
Abstract
The introduction of robotics and automation in the field of industrial production, together with the development of innovative process technolo- gies such as the deposition for additive manufacturing or spraying and automated glueing have brought to undoubted advantages and benefits in the sector, increasing the flexibility of the companies, their efficiency and ability to offer increasingly customized and de-standardized. However, the introduction of innovations in consolidated production contexts that move towards a philosophy of Industry 4.0, leads to problems and criticalities that in the past had not been considered, together with fields of research previously unexplored. What we are focusing on is the solution to the problems of Motion Planning that, compared to the past, must now con- sider contexts and constraints that in traditional machines and processes do not occur. For example, the adaptation of robots for machining and other mechanical operations, while offering great flexibility and ability to be implemented at a significantly lower cost of a machining centre, do not guarantee the dynamic performance of machines designed natively for this purpose. In addition, performances are generally not constant throughout the workspace but depend on the configuration assumed by the robot. Moreover, with regard to innovative technologies, the perfor- mances required in processes such as deposition depend on the dynamics of the deposition itself, to guarantee the quality of the final product. This introduces constraints of nature kinematic and dynamics natures in differ- ential form that need to be incorporated in Path and Trajectory Planning problem-solving methodologies. Here a method is proposed to solve the problems of Motion Planning that implements kinematic and dynamic differential constraints and provide a feasible trajectory for the machine that minimizes the geometric deviation between the proposed trajectory and the one theoretically required from the machine. Afterwards, to evaluate and have a methodology benchmark, a different approach is de- veloped by implementing Reinforcement Learning in a physical simulation environment capable to solve the Motion Planning problem using the v techniques of Artificial Intelligence. Finally, the results of the tests are discussed and the strengths are identified and compared, together with the peculiarities of the two approaches.File | Dimensione | Formato | |
---|---|---|---|
2019_10_Berardinucci.pdf
non accessibile
Descrizione: 2019_03_Berardinucci
Dimensione
6.71 MB
Formato
Adobe PDF
|
6.71 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/149613