This thesis considers the optimal control problem for multi-agent systems governed by swarmalator dynamics. The term swarmalator refers to a collective dynamical model in which each agent is given a positional state but also a phase state or oscillator state. The interactions between agents are then dictated not only by positions but also by the phase state. A wealth of global dynamical patterns arises from local interaction rules. In this thesis, we expand on the classical model of swarmalators adding the vision radius, with which it is possible to limit the number of agents that interact with each swarmalator. This guarantees a decrease in the flow of information communicated at each time step, reducing overload problems in future implementations. The collective behaviors of swarmalators that base their dynamics only on local information are similar to the one of the classical model, demonstrating that a future distributed implementation is feasible. Analysis to compare the convergence time along with the connectivity of such a system with the original model is carried out. Furthermore, we consider the swarmalator model in the context of controlled robotic swarms, that is we add a pair of control variables for each swarmalator. Then, an optimal control problem is solved which involves the minimization of the distance between the center of mass of the swarmalator system and a target position, while the dynamics are considered as equality constraints. Results show that it is possible to accurately track the centroid of the multi-agent system and steer it to any desired position in space while agents synchronize according to their parameters. In addition, we design optimal equilibrium inputs by solving the associated static optimization problem. It is then numerically shown that the controlled pattern is stable in the new equilibrium position. All numerical results have been computed using Python3 scripts and pre-built CasADi packages. Future work includes both the implementation of the system in a decentralized and distributed way and the extension of the dynamic into the three-dimensional space. This could reveal the emergence of new patterns besides making the dynamic applicable to aerial vehicles such as drones.
Questa tesi esamina il problema del controllo ottimo per sistemi multi-agente governati dalle dinamiche tipiche degli "swarmalators". Il termine swarmator si riferisce a un mod- ello dinamico collettivo in cui a ciascun agente viene assegnato uno stato posizionale ma anche uno stato di fase o oscillatorio. Le interazioni tra agenti sono quindi dettate non solo dalle posizioni ma anche dalle loro fasi. Molti pattern dinamici globali emergono dalle interazioni locali, modificando queste ultime si possono ottenere pattern differenti. In questa tesi, si espande il modello classico degli swarmalators aggiungendo il raggio di visione, con il quale è possibile limitare il numero di agenti che interagiscono con ciascuno swarmalator. Ciò garantisce una diminuzione del flusso di informazioni comunicate ad ogni step, alleggerendo il carico computazionale. I comportamenti collettivi degli swar- malators che basano le loro dinamiche solo su informazioni locali sono simili a quelli del modello classico, dimostrando la fattibilità di una futura implementazione distribuita. Viene eseguita un’analisi per confrontare il tempo di convergenza e la connettività di tale sistema al variare del raggio di visione. Inoltre, si considera il modello swarmator nel con- testo di sciami robotici controllati: una coppia di variabili di controllo è stata aggiunta per ogni swarmator. Quindi, viene risolto un problema di controllo ottimo che implica la min- imizzazione della distanza tra il centro di massa del sistema swarmator e una posizione target, mentre le dinamiche sono considerate come vincoli di uguaglianza. I risultati mostrano che è possibile seguire con precisione il centroide del sistema multi-agente e guidarlo in qualsiasi posizione desiderata nello spazio mentre gli agenti si sincronizzano in base ai loro parametri. Inoltre, vengono calcolati inputs di equilibrio ottimo risolvendo il problema di ottimizzazione statica associato. Viene quindi mostrato numericamente che il modello controllato è stabile nella nuova posizione di equilibrio. Tutti i risultati numerici sono stati calcolati utilizzando script scritti in Python3 e pacchetti CasADi. Sviluppi fu- turi includono sia l’implementazione del sistema in modo decentralizzato e distribuito che l’estensione della dinamica nello spazio tridimensionale. Ciò potrebbe rivelare l’emergere di nuovi modelli oltre a rendere la dinamica applicabile a veicoli aerei come i droni.
Optimal control of two-way coupled dynamical systems
GIACOMETTI, TOMMASO
2021/2022
Abstract
This thesis considers the optimal control problem for multi-agent systems governed by swarmalator dynamics. The term swarmalator refers to a collective dynamical model in which each agent is given a positional state but also a phase state or oscillator state. The interactions between agents are then dictated not only by positions but also by the phase state. A wealth of global dynamical patterns arises from local interaction rules. In this thesis, we expand on the classical model of swarmalators adding the vision radius, with which it is possible to limit the number of agents that interact with each swarmalator. This guarantees a decrease in the flow of information communicated at each time step, reducing overload problems in future implementations. The collective behaviors of swarmalators that base their dynamics only on local information are similar to the one of the classical model, demonstrating that a future distributed implementation is feasible. Analysis to compare the convergence time along with the connectivity of such a system with the original model is carried out. Furthermore, we consider the swarmalator model in the context of controlled robotic swarms, that is we add a pair of control variables for each swarmalator. Then, an optimal control problem is solved which involves the minimization of the distance between the center of mass of the swarmalator system and a target position, while the dynamics are considered as equality constraints. Results show that it is possible to accurately track the centroid of the multi-agent system and steer it to any desired position in space while agents synchronize according to their parameters. In addition, we design optimal equilibrium inputs by solving the associated static optimization problem. It is then numerically shown that the controlled pattern is stable in the new equilibrium position. All numerical results have been computed using Python3 scripts and pre-built CasADi packages. Future work includes both the implementation of the system in a decentralized and distributed way and the extension of the dynamic into the three-dimensional space. This could reveal the emergence of new patterns besides making the dynamic applicable to aerial vehicles such as drones.| File | Dimensione | Formato | |
|---|---|---|---|
|
2022_12_Giacometti.pdf
non accessibile
Descrizione: tesi
Dimensione
4.83 MB
Formato
Adobe PDF
|
4.83 MB | Adobe PDF | Visualizza/Apri |
|
2022_12_Giacometti_summary.pdf
non accessibile
Descrizione: executive summary
Dimensione
760.53 kB
Formato
Adobe PDF
|
760.53 kB | 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/10589/196256