Efficient Human-Robot Interaction for Robust Autonomy in Task Execution.

Robust autonomy can be achieved with learning frameworks that refine robot operating procedures through guidance from human domain experts. This work explores three capabilities required to implement efficient learning for robust autonomy: (1) identifying when to garner human input during task execution, (2) using active learning to curate what guidance is received, and (3) evaluating the tradeoff between operator availability and guidance fidelity when deciding who to enlist for guidance. We present results from completed work on interruptibility classification of collocated people that can be used to help in evaluating the tradeoff in (3).