eduzhai > Applied Sciences > Engineering >

Learning About Objects by Learning to Interact with Them

  • Save

... pages left unread,continue reading

Document pages: 17 pages

Abstract: Much of the remarkable progress in computer vision has been focused aroundfully supervised learning mechanisms relying on highly curated datasets for avariety of tasks. In contrast, humans often learn about their world with littleto no external supervision. Taking inspiration from infants learning from theirenvironment through play and interaction, we present a computational frameworkto discover objects and learn their physical properties along this paradigm ofLearning from Interaction. Our agent, when placed within the nearphoto-realistic and physics-enabled AI2-THOR environment, interacts with itsworld and learns about objects, their geometric extents and relative masses,without any external guidance. Our experiments reveal that this agent learnsefficiently and effectively; not just for objects it has interacted withbefore, but also for novel instances from seen categories as well as novelobject categories.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×