The goods delivery sector has rapidly increased in the last few years, mostly because of the changes in people habits due to COVID pandemic. Indeed, in 2020 most people experimented delivery, and a big portion still prefer it today to in person shopping. The rise in the number of parcels delivered, together with the ever increasing urbanization, gives rise to the so-called "last mile delivery problem": efficiently delivering thousands of packages to hundreds of customers in densely populated areas is a logistical nightmare, representing a significant cost in environmental, economic and traffic terms. That’s why a new frontier in goods and parcel delivery is being developed through the usage of drones, in particular ground drones, with the ability of navigation in sidewalks and pedestrian areas. Navigating in pedestrian areas means having a direct contact with people. Trying to disturb the lest possible the passers-by is one of the goals a drone should achieve in order to improve pedestrian-robot interaction and be socially accepted. A pedestrian-friendly behavior must be shown by the robot together with the ability of gracefully avoiding collisions with the obstacles it encounters while navigating. The module employed in this task is the local planner, which is responsible of deciding the control actions the robot must follow. This thesis presents a full local planning stack, which allows the robot to avoid static obstacles and pedestrians in a natural and human-friendly way. The local planner is based on the well known state of art Dynamic Window Approach, which has been modified and made aware of the people by adding a social cost function and a pedestrian simulator that allow the algorithm to consider the collaborative behavior of the pedestrians during the planning process. At the end of the development and testing process, done in a simulated environment, a real-world validation phase of the whole stack has been performed. Multiple tests with the same person or with different people, in multiple different scenarios, have been executed both with state-of-art base DWA and with modified social DWA, in order to retrieve a good, unbiased, and reliable validation of the algorithms and a comparison with a benchmark. Moreover, tests in particular situations have been performed to see the reaction of the stack in more articulated settings. Data have been analyzed using different metrics, which allowed us to look at the robot behavior from multiple perspectives. The results highlight the robustness of the proposed algorithms to real-world conditions and demonstrate the capability to perform pedestrian-friendly avoidance maneuvers with vastly superior performances with respect to the state of art DWA planner, showing that the social DWA keeps an higher safety distance and brings less disturbance to the people it encounters. Furthermore, it has shown a natural reaction to situations which could have been a big challenge.
Negli ultimi anni, soprattutto dopo la situazione mondiale dovuta al COVID che ha forzato le persone a cambiare le proprie abitudini, il settore di spedizione e consegna a domicilio ha subito un rapido incremento. Con riferimento alla consegna dell'ultimo miglio, l'aumento delle richieste, insieme all'incremento delle persone residenti in aree urbane, ha evidenziato i molti problemi legati agli attuali mezzi di trasporto. Una nuova frontiera per la consegna di pacchi e posta nelle città è in fase di sviluppo, soprattutto utilizzando droni terrestri, i quali, dovendo navigare su marciapiedi o in aree pedonali, sono a stretto contatto con i pedoni. Essere a contatto con i pedoni significa cercare di disturbarli il meno possibile in modo da avere la miglior interazione robot-pedone possibile. Un drone terrestre, mentre si muove, deve mettere proprio agio le persone che incontra e contemporaneamente evitare tutti gli ostacoli sul suo tragitto. Il modulo software che è responsabile di questo comportamento è il pianificatore locale. Questa tesi presenta un sistema completo di pianificazione locale che permette al robot di evitare ostacoli statici e pedoni in modo naturale e amichevole. Il pianificatore è basato sull'algoritmo DWA, molto noto e attuale stato dell'arte, il quale è stato modificato e reso consapevole del comportamento delle persone tramite l'aggiunta di un'apposita funzione di costo e un simulatore di pedoni che consentono al robot di pianificare la traiettoria locale tenendo in considerazione il comportamento collaborativo dei pedoni. Dopo la fase di sviluppo e di test, eseguita in un ambiente simulato, è stata effettuata una validazione sperimentale del sistema completo sul robot. Sono stati eseguiti più esperimenti con la stessa persona o con persone diverse, in scenari differenti, sia con il DWA base, attuale stato dell'arte, che con il nuovo DWA sociale, in modo da avere una validazione degli algoritmi che possa essere valida, imparziale e affidabile, e così da avere un confronto con un benchmark. In più, sono stati effettuati test in situazioni particolari in modo da investigare la reazione del robot in contesti più articolati. La versione dell'algoritmo in esecuzione sul robot è sempre stata nascosta ai pedoni testati, e, per alcuni di essi è stata la prima volta che si avvicinavano ad un robot. I dati sono stati poi analizzati utilizzando diverse metriche, le quali ci hanno consentito di osservare il comportamento del robot da più prospettive. I risultati hanno dimostrato una grande robustezza dell'algoritmo proposto agli imprevisti che possono accadere nel mondo reale e la capacità del pianificatore di avere un approccio molto gentile con i pedoni. I test hanno anche evidenziato la superiorità di questo algoritmo nell'affrontare i pedoni rispetto al DWA base, attuale stato dell'arte, mostrando come il DWA sociale mantenga una maggiore distanza di sicurezza e porti meno disturbo quando incontra delle persone durante la navigazione.
Design, development and experimental validation of a socially aware local planner for a mobile robot
Portanti, Samuele
2021/2022
Abstract
The goods delivery sector has rapidly increased in the last few years, mostly because of the changes in people habits due to COVID pandemic. Indeed, in 2020 most people experimented delivery, and a big portion still prefer it today to in person shopping. The rise in the number of parcels delivered, together with the ever increasing urbanization, gives rise to the so-called "last mile delivery problem": efficiently delivering thousands of packages to hundreds of customers in densely populated areas is a logistical nightmare, representing a significant cost in environmental, economic and traffic terms. That’s why a new frontier in goods and parcel delivery is being developed through the usage of drones, in particular ground drones, with the ability of navigation in sidewalks and pedestrian areas. Navigating in pedestrian areas means having a direct contact with people. Trying to disturb the lest possible the passers-by is one of the goals a drone should achieve in order to improve pedestrian-robot interaction and be socially accepted. A pedestrian-friendly behavior must be shown by the robot together with the ability of gracefully avoiding collisions with the obstacles it encounters while navigating. The module employed in this task is the local planner, which is responsible of deciding the control actions the robot must follow. This thesis presents a full local planning stack, which allows the robot to avoid static obstacles and pedestrians in a natural and human-friendly way. The local planner is based on the well known state of art Dynamic Window Approach, which has been modified and made aware of the people by adding a social cost function and a pedestrian simulator that allow the algorithm to consider the collaborative behavior of the pedestrians during the planning process. At the end of the development and testing process, done in a simulated environment, a real-world validation phase of the whole stack has been performed. Multiple tests with the same person or with different people, in multiple different scenarios, have been executed both with state-of-art base DWA and with modified social DWA, in order to retrieve a good, unbiased, and reliable validation of the algorithms and a comparison with a benchmark. Moreover, tests in particular situations have been performed to see the reaction of the stack in more articulated settings. Data have been analyzed using different metrics, which allowed us to look at the robot behavior from multiple perspectives. The results highlight the robustness of the proposed algorithms to real-world conditions and demonstrate the capability to perform pedestrian-friendly avoidance maneuvers with vastly superior performances with respect to the state of art DWA planner, showing that the social DWA keeps an higher safety distance and brings less disturbance to the people it encounters. Furthermore, it has shown a natural reaction to situations which could have been a big challenge.File | Dimensione | Formato | |
---|---|---|---|
Thesis_Samuele_Portanti.pdf
non accessibile
Descrizione: Thesis
Dimensione
13.7 MB
Formato
Adobe PDF
|
13.7 MB | Adobe PDF | Visualizza/Apri |
Executive_Summary_Thesis_Samuele_Portanti.pdf
non accessibile
Descrizione: Executive Summary
Dimensione
1.48 MB
Formato
Adobe PDF
|
1.48 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/10589/196991