KR$^3$: An Architecture for Knowledge Representation and Reasoning in\n Robotics

Abstract

This paper describes an architecture that combines the complementary\nstrengths of declarative programming and probabilistic graphical models to\nenable robots to represent, reason with, and learn from, qualitative and\nquantitative descriptions of uncertainty and knowledge. An action language is\nused for the low-level (LL) and high-level (HL) system descriptions in the\narchitecture, and the definition of recorded histories in the HL is expanded to\nallow prioritized defaults. For any given goal, tentative plans created in the\nHL using default knowledge and commonsense reasoning are implemented in the LL\nusing probabilistic algorithms, with the corresponding observations used to\nupdate the HL history. Tight coupling between the two levels enables automatic\nselection of relevant variables and generation of suitable action policies in\nthe LL for each HL action, and supports reasoning with violation of defaults,\nnoisy observations and unreliable actions in large and complex domains. The\narchitecture is evaluated in simulation and on physical robots transporting\nobjects in indoor domains; the benefit on robots is a reduction in task\nexecution time of 39% compared with a purely probabilistic, but still\nhierarchical, approach.\n