Abstract
In this paper, we present an attention mechanism for mobile robots to face the problem of place categorization. Our approach, which is based on active perception, aims to capture images with characteristic or distinctive details of the environment that can be exploited to improve the efficiency (quickness and accuracy) of the place categorization. To do so, at each time moment, our proposal selects the most informative view by controlling the line-of-sight of the robot’s camera through a pan-only unit. We root our proposal on an information maximization scheme, formalized as a next-best-view problem through a Markov Decision Process (MDP) model. The latter exploits the short-time estimated navigation path of the robot to anticipate the next robot’s movements and make consistent decisions. We demonstrate over two datasets, with simulated and real data, that our proposal generalizes well for the two main paradigms of place categorization (object-based and image-based), outperforming typical camera-configurations (fixed and continuously-rotating) and a pure-exploratory approach, both in quickness and accuracy.
Highlights
Mobile robots are increasingly gaining presence in humancentered environments, like houses [1,2] or convention centers [3,4]
In this work we focus on the problem of semantic place categorization, which refers to the problem of assigning a semantic label to places or parts of the environment once their geometry is already known [12]
We propose a probabilistic framework built upon an information maximization scheme formalized as a next-best-view problem, as well as on Markov Decision Processes (MDP) to exploit the expected short-term robot path and the previously gathered semantic knowledge
Summary
Mobile robots are increasingly gaining presence in humancentered environments, like houses [1,2] or convention centers [3,4]. The video acquired by a mobile robot with a fixed on-board camera will probably include many non-informative images, since the line-of-sight is fixed, typically along the tangent of the robot path This has an important impact on the categorization efficiency by constraining the observed areas of the environment, being necessary to either adapt the robot’s path during the inspection [14], or to employ advanced image representations and temporal fusion methods to overcome it. To select the line-of-sight, we exploit knowledge about the robot pose, the short-time estimated navigation path, the camera parameters and the environment geometry, including the segmentation of the different rooms or spaces in it The latter is a common assumption in place categorization problems where the goal is to determine the most probable label for each segmented space in the environment [15].
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
