Embodied queries for robot task learning

Personal robots are intended to assist people in their daily lives; therefore, it is important for them to efficiently learn how to accomplish tasks delineated by their human partners, in the environment in which they are situated. Toward that end however, it is arguably even more important to enable these robots to characterize their own uncertainty as they attempt to generalize task-specific knowledge learned and equip them with a domain-independent framework for asking the appropriate questions of their human partners to resolve the uncertainty. The work presented in this extended abstract is a high level overview of current and future work that we are developing in this domain, using learning from demonstration and active robot learning.