This website uses browsing/session and functional cookies to ensure you get the best experience. Learn More
Title: An Approach to Self-Extension: Enabling a Mobile Robot to Reason about Different Epistemic Actions and Uncertainty for Autonomous Knowledge Gathering
Abstract: In any real world task a robot tries to accomplish, it faces two significant challenges that it needs to deal with:
(i) its knowledge about the world is incomplete, so substantial knowledge required to successfully achieve a given goal is missing; and (ii) the knowledge it might have about the world is uncertain, due to noise in sensing and/or unreliability of accessible knowledge sources.
Nevertheless, we expect our robot to behave intelligently to achieve the given goal robustly and efficiently. In this talk I am presenting Dora, our integrated robot systems that explores its environment in a task-driven way to address those two challenges and extend its knowledge about the world autonomously. It can accomplish a variety of different epistemic goals in a real world full of uncertainties employing a probabilistic approach to representation, reasoning and planning. It integrates and plans to gather evidence obtained through computer vision and pattern recognition algorithms (object detection, room categorisation), from interaction with humans, and from exploiting common-sense knowledge queried from the WWW. By intelligently combining those knowledge sources Dora can accomplish a variety of different epistemic goals in order to (self-)extend its knowledge about the world. In order to generate goal-driven behaviour Dora features a novel switching planner that allows the system to schedule actions implemented by a number of competences that gather knowledge from the various knowledge sources. I will present the different competences implemented in our system, the systemic probabilistic approach to plan to perceive, taking into account the uncertainties, and the overall architecture. We will see how Dora autonomously explores an unknown map; how it learns about the category of rooms in this map (e.g. kitchen, corridor, etc.); and how it autonomously determines the location of a specified object. A predecessor to this system was nominated for the IJCAI 2011 video award: