Autonomous Discovery of Abstract Notions by a Robot#

Ivan Bratko
University of Ljubljana, Slovenia


How could a robot, on its own, possibly discover abstract notions such as a general concept of a tool? In this talk, I will describe one possible approach to this, and experiments in autonomous discovery of abstract concepts in a robotic domain. An autonomous robot is performing tasks in its world, collecting data and learning predictive theories about the world. In particular, we are interested in the robot's “gaining insights”, by inventing new abstract concepts that enable the simplification of the robot's current theory about the robot’s world. Examples of insights are discoveries of concepts like mobility, obstacle, stability, etc. It should be noted that these concepts are not explicitly present in the robot’s sensory observations. A particular challenge is to discover functional roles of objects in solving robot manipulation tasks. In experiments in robot’s learning from its plans to solve concrete tasks, the concepts of a tool and obstacle emerged. Our approach employs Inductive Logic Programming with predicate invention.

Imprint Privacy policy « This page (revision-5) was last updated on Tuesday, 5. August 2014, 14:54 by Kaiser Dana
  • operated by