A robot rehearses internally and learns an affordance relation

This paper introduces a novel approach to a crucial problem in robotics: Constructing robots that can learn general affordance relations from their experiences. Our approach has two components. (a) The robot models affordances as statistical relations between actual actions, object properties and the experienced effects of actions on objects. (b) To exploit the general-knowledge potential of its actual experiences, the robot, much like people, engages in internal rehearsal, playing-out ldquoimaginedrdquo scenarios grounded in but different from actual experience. To the extent the robot veridically appreciates affordance relations, the robot can autonomously predict the outcomes of its behaviors before executing them. Accurate outcome prediction in turn facilitates planning of a sequence of behaviors, toward executing the robotpsilas given task successfully. In this paper, we report very first steps in this approach to affordance learning, viz., the results of simulations and humanoid-robot-embodied experiments targeted toward having the robot learn one of the simplest of affordance relations, that a space affords traversability vs. impediment to a goal-object in the space.