Representing and Reasoning with Intentional Actions on a Robot

This paper describes a general architecture for robots to represent and reason with intentional actions. The architecture reasons with tightly-coupled transition diagrams of the domain at two different resolutions. Non-monotonic logical reasoning with a coarse-resolution transition diagram is used to compute a plan comprising intentional abstract actions for any given goal. Each such abstract action is implemented as a sequence of concrete actions by reasoning over the relevant part of the fine-resolution transition diagram, with the outcomes of probabilistic execution of the concrete actions being added to the coarse-resolution history. The capabilities of this architecture are illustrated in the context of a simulated robot assisting humans in an office domain, on a physical robot (Baxter) manipulating tabletop objects, and on a wheeled robot (Turtlebot) moving objects to particular places or people in an office. We show that this architecture improves reliability and efficiency in comparison with a planning architecture that does not include intentional actions.