Context-sensitive in-the-world interfaces for mobile manipulation robots

We present an interface that allows users to direct a mobile manipulation robot in tabletop pick-and-place tasks using only their head motions and a single button. The system uses an estimate of the user's head pose and a 3d world model maintained by the robot to determine where the user is pointing their head. We give the results of some preliminary evaluations of our system, which suggest that it is both intuitive and effective. We also describe an example trash-sorting application where the user directs a PR2 robot sort objects in to “trash” and “recycle” piles.