Advancing Socially Aware Navigation
Loading...
Authors
Salek Shahrezaie, Roya
Issue Date
2024
Type
Dissertation
Language
en_US
Keywords
Social Intelligence In Robotics , Socially Aware Navigation
Alternative Title
Abstract
Robots have evolved from their traditional roles in manufacturing, especially in the automotive sector, where they have been used for decades. Today, they are increasingly integrated into everyday life services, transforming societies and public spaces. This trend is expected to continue as technological advancements improve robot capabilities and labor shortages grow in developed countries. With their versatile functions, robots are now used in various public service contexts. They provide exhibit information, conduct virtual tours in museums and galleries, disinfect areas, serve as security guards, and engage in educational activities for visitors, including schoolchildren. Regardless of their role, the challenge remains the same: navigating efficiently, reliably, and appropriately around people in shared social environments. For robots to be accepted in these environments, it is crucial to develop navigation strategies tailored to the social contexts of each setting. While collision-free navigation in dynamic environments is a solved problem, navigating in human-robot environments presents new challenges. The focus has shifted from merely moving efficiently from one point to another to autonomously detecting the context and adapting to appropriate social navigation strategies. Museums, with their diverse navigation contexts in a compact space, are ideal for studying such behaviors. Our previous socially aware navigation model involved context classification, object detection, and predefined rules for navigation behavior in contexts like hallways or queues. This work extends these concepts by incorporating environmental context, object information, and more realistic interaction rules for complex social spaces. In the project's initial phase, we translated real-world interactions into rules for robotic navigation systems. Additionally, we use context recognition, object detection, and scene data to select context-appropriate rules. We present our methodology for studying social behaviors in complex contexts, various analyses of our text corpus for museums, and extracting social norms. We present scenarios where robot navigation is different based on the context. Finally, we present algorithms and implementation techniques and demonstrate the application of some of these rules in simulated scenarios and through a real-world demonstration using a Ridgeback robot. This approach aims to enhance the long-term applicability of social robots in dense human environments by ensuring they navigate in a socially aware manner.
Description
Citation
Publisher
License
CC BY