http://www.BlabDroid.com
boxie.media.mit.edu
Documentaries are typically captured in a very structured way,
using teams to film and interview people. We developed an
autonomous method for capturing structured documentaries
through an interactive robotic camera, which was used as a
mobile physical agent to facilitate interaction and story gathering
within a ubiquitous media framework. We sent this robot out to
autonomously gather human narrative about its environment. The
robot had a specific story capture goal and leveraged humans to
attain that goal. The robot collected a 1st person view of stories
unfolding in real life, and as it engaged with its subjects via a
preset dialog, these media clips were intrinsically structured.
We evaluated this agent by way of determining "complete"
vs. "incomplete" interactions. "Complete" interactions were
those that generated viable and interesting videos, which could
be edited together into a larger narrative. It was found that 30%
of the interactions captured were "complete" interactions. Our
results suggested that changes in the system would only produce
incrementally more "complete" interactions, as external factors
like natural bias or busyness of the user come into play. The types
of users who encountered the robot were fairly polar; either they
wanted to interact or did not - very few partial interactions went
on for more than 1 minute. Users who partially interacted with the
robot were found to treat it rougher than those who completed the
full interaction. It was also determined that this type of limited-
interaction system is best suited for short-term encounters. At the
end of the study, a short documentary showcasing the people and
activity in our building was easily produced from the structured
videos that were captured, indicating the utility of this approach.
