Among the different strategies used to prevent a humanoid robot from falling, when external disturbances are applied, the most effective one is the stepping strategy. The aim of this thesis is therefore, to contribute towards the implementation of such capabilities on the TORO robot, a humanoid platform at the German Aerospace Centre. This work presents the capturability-based analysis as a definition of stability during stepping, which is also used for selecting the number of steps necessary for the push recovery. Various analysis and comparisons are done on simplified models for single step recovery strategies. A hierarchical approach is followed towards selecting the most convenient push recovery approach. Models and in-depth analysis are therefore presented for triggering the step, deciding which stepping strategy is necessary and online generation of the trajectories through Model Predictive Control and the Divergent Component of Motion methods. Then, a passivity-based full body controller is presented capable of following the desired reference trajectories. The previous methods are applied and evaluated, together with the physical model of the robot in the OpenHRP simulation environment. Finally, considerations are made on the results obtained through simulation and their applicability on the real hardware.
Tra le varie strategie che un robot umanoide, in caso di spinte esterne, può mettere in atto per evitare una caduta, la più efficace è compiere un passo. Questa tesi si concentra perciò sull’implementazione e lo studio di tale strategia basandosi sulla piattaforma TORO, un robot umanoide sviluppato presso l’Agenzia Aerospaziale Tedesca. Viene presentato il concetto di catturabilità come misura di stabilità durante il compimento del passo. Inoltre, vengono analizzati e messi a confronto diversi modelli semplificati per il recupero dell’equilibrio attraverso il passo in avanti. A questo proposito, il robot segue un percorso di decisione gerarchico che lo porta a scegliere quale strategia utilizzare. Per questo vengono trattati ed analizzati gli algoritmi attraverso i quali il robot stabilisce se compiere o meno il passo, sceglie dove posizionare il piede una volta deciso di fare il passo e infine quelli attraverso i quali calcola le traiettorie per gli end-effectors principali. Quest’ultima operazione viene fatta attraverso gli algoritmi di Model Predictive Control e Divergent Component of Motion. Viene inoltre descritto il controllore utilizzato sul robot TORO. Infine, i risultati ottenuti vengono valutati attraverso la simulazione del comportamento del modello sul software OpenHRP. La tesi si conclude con le valutazioni circa la fattibilità dell’implementazione sull’hardware reale.
Comparative analysis of push recovery strategies for humanoid robots
BOGDANOVSKI, DANIEL
2016/2017
Abstract
Among the different strategies used to prevent a humanoid robot from falling, when external disturbances are applied, the most effective one is the stepping strategy. The aim of this thesis is therefore, to contribute towards the implementation of such capabilities on the TORO robot, a humanoid platform at the German Aerospace Centre. This work presents the capturability-based analysis as a definition of stability during stepping, which is also used for selecting the number of steps necessary for the push recovery. Various analysis and comparisons are done on simplified models for single step recovery strategies. A hierarchical approach is followed towards selecting the most convenient push recovery approach. Models and in-depth analysis are therefore presented for triggering the step, deciding which stepping strategy is necessary and online generation of the trajectories through Model Predictive Control and the Divergent Component of Motion methods. Then, a passivity-based full body controller is presented capable of following the desired reference trajectories. The previous methods are applied and evaluated, together with the physical model of the robot in the OpenHRP simulation environment. Finally, considerations are made on the results obtained through simulation and their applicability on the real hardware.File | Dimensione | Formato | |
---|---|---|---|
2018_04_Bogdanovski.pdf
non accessibile
Descrizione: Testo della Tesi
Dimensione
13.69 MB
Formato
Adobe PDF
|
13.69 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/10589/140042