IEEE Robotics & Automation Magazine - March 2012 - 109

FROM THE FIELD

Taking the Long View: A Report on Two Recent
Workshops on Long-Term Autonomy
By Jonathan Kelly, Gabe Sibley, Tim Barfoot, and Paul Newman

he topics of long-term autonomy and lifelong learning
are attracting an increasing
amount of attention in the
robotics research community. Robots
now routinely operate without human
intervention for short periods of time,
and several systems have demonstrated operation over longer durations. However, in the future, the
majority of robots will need to function autonomously, outside the laboratory or the factory, on time scales
ranging from days to years. Developing adaptive and flexible machines
capable of this level of independence will require significant technical advances.
In this short report, we summarize the results of two recent
workshops on long-term autonomy
and lifelong learning: 1) the International Conference on Robotics
and Automation (ICRA) 2011 Workshop on Long-Term Autonomy
and 2) the Robotics Science and
Systems (RSS) 2011 Workshop on
Autonomous Long-Term Operation in Novel Environments
(ALONE). These workshops brought
together prominent researchers from
a diverse range of subdisciplines to
establish the key issues involved in
building robotic systems that are
able to operate in increasingly
large-scale environments and over
long periods of time. Many of the
problems identified are difficult. At

Digital Object Identifier 10.1109/MRA.2011.2181792
Date of publication: 26 March 2012

present, we can offer numerous questions but few answers, and this, in a
way, is the defining characteristic of a
new research area. Based on the success of the initial discussions, longterm autonomy clearly warrants an
extended dialogue, with the next step
being a follow-up workshop at ICRA
2012.

(DARPA) Grand Challenge, the
Mars Exploration Rovers, Willow
Garage PR2 demos, and Atlantic
Ocean-crossing autonomous underwater vehicles (AUVs) are only
some examples of the increasing
maturity of the field as a whole.
However, with these successes, new
challenges related to robust longterm operation and lifelong learnChallenges for Persistent
ing are also emerging.
Robotic Systems
In many respects, the problem
State-of-the-art results in mapping domain is unlike any other. Particuand teach-and-repeat systems have larly, long-term experiments often
convincingly shown that robots are require a level of systems integration
able to operate over substantial spatial substantially above that of experiand temporal extents. The Defense ments that test only one or two priAdvanced Research Projects Agency mary technologies. This integration
bar must be surmounted before
the actual research can begin.
There is an additional complication of academic verification-
it can be very difficult to repeat
another group's work to replicate their results. Simply replaying
logged data is straightforward, but
online decision-making and reactive control are substantially
more complex to implement.
Both the ICRA and RSS workshops dealt with a broad range of
subjects related to the construction of reliable, persistent autonomous systems. This encompassed
long-term planning and exploration strategies; long-term adaptation; estimation in dynamic
environments; perceptual representations incorporating space and
time; exploitation of semantic
Bowler WildCat autonomous vehicle platform.
information; semiautonomy and
(Photo courtesy of the University of Oxford Mobile
human-in-the-loop control; fault
Robotics Group.)
MARCH 2012

IEEE ROBOTICS & AUTOMATION MAGAZINE

109

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