IEEE Robotics & Automation Magazine - March 2012 - 110

Field robot on Devon Island in 2009 testing long-range autonomous driving in
planetary analog domain. (Photo courtesy of Tim Barfoot.)
l

tolerance and failure prediction; novel
sensors, actuators, and power sources;
resource-constrained operation; robust
systems engineering; and online calibration and identification.
One immediate task is to determine how to best ensure the future
success of the subfield. Can we ascertain where researchers should focus
their efforts now? Can we ascertain
where researchers should focus their
efforts now, to further facilitate the
development and
* deployment of
persistent systems?
The topics of
The community
long-term autonomy
has started to consider ways to exand lifelong learning
tend safe and
are attracting an
reliable operation
for longer duraincreasing amount of
tions. What valuattention in the
able lessons have
we learned from
robotics research
large-scale robotic
community.
experiments? What
* are the fundamental obstacles
that need to be overcome to ensure
robust and continual operations?
What is the most suitable way to
move current research from the laboratory into long-term service
roles? These problems will be addressed as part of the ongoing
110

IEEE ROBOTICS & AUTOMATION MAGAZINE

discourse that has been initiated by
the workshops.
Outcomes
Several common themes emerged
from the workshop discussions,
beyond simply establishing that the
problems are especially hard in general. The themes included, e.g., the
importance of redundant sensing for
robustness, the need for novel map
representations to improve large-scale
navigation, the use of semiautonomy
as a stepping stone to speed progress,
and the importance of long-duration
field experiments to uncover system
weaknesses.
Another outcome was a list of
pertinent questions, whose answers
will help to guide the evolution of the
subfield. Here, we outline these questions and give some related context.
l What is long-term autonomy? The
phrase "long-term autonomy" is
inherently broad and can often
mean different things to different
people. A concise definition and
appropriate scoping may be necessary to secure funding from government agencies and other
groups.
l What characterizes successful longterm autonomy? Related to the
previous question, how does one
characterize the success of these

MARCH 2012

systems? What metrics are available? Are metrics specific to individual projects, or can we say
something more generally? How
can we evaluate our progress?
How far will sensing technology
get us? What new sensors are
required? Many problems that are
algorithmically difficult can be
substantially mitigated or solved
through improvements in sensing. Indeed, the release of products, such as the Velodyne laser
scanner and the Microsoft Kinect
RGB-D sensor, has largely revolutionized simultaneous localization
and mapping (SLAM) research.
What sensing advances can we
depend on? And what types of new
sensors would be most useful?
What is certification? What does it
mean to certify these systems as
safe? What constitutes a safe system? What process will be required
to certify a system as safe for longterm operation? What steps will be
necessary for public acceptance and
to ensure public trust?
Are these problems appropriate for
the academic world? Is long-term
autonomy primarily an engineering
problem? Or are there deep academic questions to be answered?
What role should academia play in
developing these systems?
What is an acceptable level of
human intervention or human-inthe-loop control? Some systems will
not be fully autonomous but will
instead rely on a human in the
loop. What amount of human
involvement is acceptable? Can
some degree of human-in-theloop control reduce the risks
involved in development and
deployment? Can human involvement be used as a stepping
stone to get systems up and running
more quickly with dependence then
reduced over time?
Is it too soon to ask these questions?
Early in the proposal phase of both
workshops, a point was made that
it may be too soon to ask these
questions. Has robotics advanced
to the stage where the fielding of

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - March 2012