Self-assessing and communicating manipulation proficiency through active uncertainty characterization

Abstract

Autonomous manipulation has the potential to improve the quality of life of many by assisting in routine household tasks such as cooking, cleaning, and organizing. However, for safe, dependable, and effective operation alongside humans, both the robot and the human must have an accurate and reliable assessment of the robot's proficiency at completing the relevant tasks. Such an assessment helps to ensure that the robot does not engage in tasks that it cannot handle and instead engages in tasks that are well-aligned with the robot's abilities. This proposal thus investigates how a robot can actively assess both its proficiency and its confidence in that assessment through appropriate measures of uncertainty that can be efficiently and effectively communicated to a human. The experiments examine how a user's trust and subsequent use of a robot vary as a result of the robot's self-assessment of proficiency.