17-03-2014, 10:29 AM
ABSTRACT
We present a 3D referencing technique tailored for remote
maintenance tasks in augmented reality. The goal is to improve
the accuracy and efficiency with which a remote expert can point
out a real physical object at a local site to a technician at that site.
In a typical referencing task, the remote expert instructs the local
technician to navigate to a location from which a target object can
be viewed, and then to attend to that object. The expert and technician
both wear head-tracked, stereo, see-through, head-worn
displays, and the expert’s hands are tracked by a set of depth cameras.
The remote expert first selects one of a set of prerecorded
viewpoints of the local site, and a representation of that viewpoint
is presented to the technician to help them navigate to the correct
position and orientation. The expert then uses hand gestures to
indicate the target.
INTRODUCTION
For participants to collaborate successfully on a spatial/action
task, they must be able to generate and interpret effective reference
cues. Successful spatial/action coordination is challenged
when participants are not colocated and cannot directly gesture in
each other’s environments. When spatial communication relies
only on language, errors occur even when communication partners
think they have agreed [5].
Spatial/action coordination is critical in remote task assistance
in which an expert assists someone with less knowledge in that
domain (e.g., a technician, in an equipment maintenance domain)
and both are in physically separate environments. The remote
expert must explain to the technician task procedures involving
objects in the task space local to the technician who performs the
procedures. Thus, the efficiency and effectiveness of remote task
assistance depends heavily on the referencing techniques used.
Referencing in remote task assistance involves two successive
steps: localization and selection. In localization, the expert must
help the technician navigate to a destination location in the task
space where a target object can be clearly seen by the technician.
(Note that this step may not be necessary if the task environment
is sufficiently simple.) This process can be particularly challenging
if the task space is complicated and the target is visible only
from a small set of head positions and view angles. Once the target
is within the technician’s view, selection occurs when the
expert indicates the target to the technician (e.g., by pointing at it).
To address these problems, we are developing an augmented
reality referencing technique (Figure 1), based on 3D gestural
interaction. The technique is intended for situations in which both
the expert and technician wear head-tracked, stereo, see-through,
head-worn displays (HWDs), the technician’s head-worn stereo
camera views are transmitted to the expert, and the expert’s hands
are tracked. Our initial task domain is aircraft engine maintenance.