With the rapid growth of e-commerce in recent years, demand for fast and reliable delivery services has drastically increased. However, in the current logistics model, a significant part of the economic and environmental costs refers to the "last-mile delivery", which is a delivery process from the city logistic center to the final recipient. Therefore, parcel delivery by autonomous vehicles is the ultimate goal of the cheap and sustainable delivery process. YAPE, the autonomous mobile robot specifically designed for this purpose, is in the focus of this thesis, completed with the mOve research group from Politecnico di Milano. Moving in urban environments requires the "social intelligence" of the robot to be accepted and integrated into human crowded areas while achieving desired goals. However, robots usually fail to find a safe path in crowded environments due to the high uncertainty of future pedestrians' trajectories caused by the variability of pedestrians’ behavior. This thesis presents an extension of the "Dynamic Window Approach" algorithm that enables the robot to understand the human interaction policies through human motion prediction. Accurate prediction of the future pedestrians' trajectories is achieved by using the Social Force Model. The algorithm was implemented within the ROS move_base navigation stack and was tested in an ad-hoc designed simulation environment. The developed algorithm has shown considerably better results than the original version in most situations commonly seen in pedestrian traffic. Most notably, the robot's influence on pedestrians is significantly reduced by the comprehensive and compact term introduced in the multi-objective cost function, called social cost. Furthermore, the algorithm yields high robustness regarding the critical points of the human motion prediction, which are: (1) intrinsic differences in pedestrian behavior, (2) errors in the prediction of the pedestrian goal, and (3) pedestrian desired speed. Finally, it allows straightforward integration in the existing navigation stack. In conclusion, the developed algorithm has great potential to substantially improve YAPE navigation in crowded environments.
Con la rapida crescita dell’e-commerce degli ultimi anni è aumentata con decisione la domanda per servizi di spedizione veloci e affidabili. Nell'attuale modello logistico una parte di significativa dei costi, sia economici che ambientali, è concentrata nell’ “ultimo miglio”, ovvero nella spedizione dall’hub logistico cittadino al del destinatario finale. Per questo motivo robot mobili autonomi sono sperimentati da numerose aziende nel campo delle consegne in ambiente urbano. YAPE, il robot oggetto di questa tesi, è stato progettato specificatamente con questo scopo. Per essere in grado di raggiungere in completa autonomia il destinatario di una spedizione, però, YAPE deve essere in grado di convivere e interagire con i pedoni incontrati durante la navigazione. In ambienti affollati i classici algoritmi di navigazione faticano a trovare una traiettoria senza collisioni, a causa dell’alta variabilità dei comportamenti umani e della conseguente difficoltà nel prevedere la futura traiettoria di un passante. Questa tesi presenta un’estensione dell’algoritmo “Dynamic Window Approach” che consente al robot di predire la traiettoria futura dei pedoni. La predizione è effettuata attraverso l’utilizzo del modello di interazione “Social Forces Model”, che cattura realisticamente il comportamento umano in ambienti con più agenti. L’algoritmo è stato implementato all'interno dello stack di navigazione ROS move_base ed è stato testato in condizioni tipiche del traffico pedonale, in un ambiente di simulazione progettato ad-hoc. Le simulazioni mostrano la superiorità dell’approccio presentato rispetto all'algoritmo di navigazione originario. Inoltre, gli esperimenti effettuati dimostrano che i risultati sono robusti a incertezze sul comportamento dei pedoni (livello di confidenza, velocità e obiettivo del pedone).
Design and development of socially aware navigation for a mobile robot
KNEZEVIC, PETAR
2020/2021
Abstract
With the rapid growth of e-commerce in recent years, demand for fast and reliable delivery services has drastically increased. However, in the current logistics model, a significant part of the economic and environmental costs refers to the "last-mile delivery", which is a delivery process from the city logistic center to the final recipient. Therefore, parcel delivery by autonomous vehicles is the ultimate goal of the cheap and sustainable delivery process. YAPE, the autonomous mobile robot specifically designed for this purpose, is in the focus of this thesis, completed with the mOve research group from Politecnico di Milano. Moving in urban environments requires the "social intelligence" of the robot to be accepted and integrated into human crowded areas while achieving desired goals. However, robots usually fail to find a safe path in crowded environments due to the high uncertainty of future pedestrians' trajectories caused by the variability of pedestrians’ behavior. This thesis presents an extension of the "Dynamic Window Approach" algorithm that enables the robot to understand the human interaction policies through human motion prediction. Accurate prediction of the future pedestrians' trajectories is achieved by using the Social Force Model. The algorithm was implemented within the ROS move_base navigation stack and was tested in an ad-hoc designed simulation environment. The developed algorithm has shown considerably better results than the original version in most situations commonly seen in pedestrian traffic. Most notably, the robot's influence on pedestrians is significantly reduced by the comprehensive and compact term introduced in the multi-objective cost function, called social cost. Furthermore, the algorithm yields high robustness regarding the critical points of the human motion prediction, which are: (1) intrinsic differences in pedestrian behavior, (2) errors in the prediction of the pedestrian goal, and (3) pedestrian desired speed. Finally, it allows straightforward integration in the existing navigation stack. In conclusion, the developed algorithm has great potential to substantially improve YAPE navigation in crowded environments.File | Dimensione | Formato | |
---|---|---|---|
2021_Thesis_Petar_Knezevic.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
33.78 MB
Formato
Adobe PDF
|
33.78 MB | Adobe PDF | Visualizza/Apri |
2021_Executive_Summary_Petar_Knezevic.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
301.88 kB
Formato
Adobe PDF
|
301.88 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/10589/183350