IEEE Robotics & Automation Magazine - December 2022 - 156

latency compensation for various high
speeds. The UT Austin team was the
only team that explicitly considered
latency compensation in its AMRL stack,
through a latency parameter. During
high-speed maneuvers, the robot inevitably
needs to aggressively change its
navigation speed to swerve around
obstacles and accelerate in open spaces.
System latency caused by sensing, processing,
computation, communication,
and actuation will likely invalidate previously
feasible plans. While simply
tuning the latency parameter value can
help to certain extent, a more intelligent
and adaptive way to calculate and compensate
system latency is necessary for
the robot to take full advantage of its
computing power before executing
aggressive maneuvers.
Navigation Is More
Than Planning
To plan agile navigation maneuvers
through highly constrained obstacle
environments, the robot first needs to
accurately perceive its configuration
with respect to the obstacles. Inaccurate
localization and odometry during fast
maneuvers with significant angular
velocity usually produce significant
drift, causing previously valid plans to
become infeasible. While all three
teams' local planners rely on raw perception
to minimize such an adverse
effect, e.g., by using high-frequency
laser scans and directly planning with
respect to these raw features, their
global planner usually depends on the
results of localization and odometry
techniques. For example, the Temple
team used Dijkstra's global planner in
move_base. An erroneous localization
will cause an erroneous global plan,
which, in turn, will affect the quality of
the local plan.
Such an adverse effect will diminish
when the navigation speed is low
because localization techniques may
recover from drift over time. During
high-speed navigation, however, the
planner needs to quickly plan actions
regardless of whether the drift has been
fixed or not. As mentioned previously,
latency will start to play a role, as well,
because a good latency compensation
technique will depend on an accurate
localization and odometry model of the
robot, i.e., being able to predict where
the robot will be based on where the
robot is and what action will be executed.
Techniques for better odometry,
localization, and kinodynamic models
during high-speed navigation will be
necessary to allow mobile robots to
move both fast and accurately at the
same time.
Conclusions
The results of the BARN Challenge at
ICRA 2022 suggest that, contrary to
the perception of many in the field,
autonomous metric ground robot navigation
cannot yet be considered a
solved problem. Indeed, even the competition
organizers initially assumed
that obstacle avoidance alone was too
simple a goal and therefore emphasized
navigation speed before the physical
competition. However, each of the
finalist teams experienced difficulty
performing collision-free navigation,
and this ultimately led the organizers
to modify the competition rules to
focus more on collision avoidance.
This result suggests that state-of-theart
navigation systems still suffer from
suboptimal performance potentially
due to many aspects of the full navigation
system (as discussed in the " Discussions "
section). Therefore, while it
is worthwhile to extend navigation
research in directions orthogonal to
metric navigation (e.g., purely visionbased,
off-road, and social navigation),
the community should not overlook
the problems that still remain in this
space, especially when robots are
expected to be extensively and reliably
deployed in the real world.
Xuesu Xiao, George Mason University
and Everyday Robots, Fairfax, VA
22030 USA. E-mail: xiao@gmu.edu.
Zifan Xu, The University of Texas at
Austin, Austin, TX 78712 USA. E-mail:
zfxu@utexas.edu.
Zizhao Wang, The University of Texas at
Austin, Austin, TX 78712 USA. E-mail:
zizhao.wang@utexas.edu.
156 * IEEE ROBOTICS & AUTOMATION MAGAZINE * DECEMBER 2022
Yunlong Song, University of Zurich,
Zurich 8050, Switzerland. E-mail:
song@ifi.uzh.ch.
Garrett Warnell, Army Research Laboratory
and The University of Texas at
Austin, Austin, TX 78712 USA. E-mail:
garrett.a.warnell.civ@mail.mil.
Peter Stone, The University of Texas at
Austin and Sony AI, Austin, TX 78712
USA. E-mail: pstone@cs.utexas.edu.
Tingnan Zhang, Robotics@Google,
Mountain View, CA 94043, USA.
E-mail: tingnan@google.com.
Shravan Ravi, The University of Texas at
Austin, Austin, TX 78712 USA. E-mail:
shravan.ravi@utexas.edu.
Gary Wang, The University of Texas at
Austin, Austin, TX 78712 USA. E-mail:
gary_wang@utexas.edu.
Haresh Karnan, The University of Texas
at Austin, Austin, TX 78712 USA.
E-mail: haresh.miriyala@utexas.edu.
Joydeep Biswas, The University of Texas
at Austin, Austin, TX, 78712 USA.
E-mail: joydeepb@cs.utexas.edu.
Nicholas Mohammad, University of Virginia,
Charlottesville, VA 22903 USA.
E-mail: nm9ur@virginia.edu.
Lauren Bramblett, University of Virginia,
Charlottesville, VA 22903 USA. E-mail:
qbr5kx@virginia.edu.
Rahul Peddi, University of Virginia,
Charlottesville, VA 22903 USA.
E-mail: rp3cy@virginia.edu.
Nicola Bezzo, University of Virginia,
Charlottesville, VA 22903 USA.
E-mail: nbezzo@virginia.edu.
Zhanteng Xie, Temple University, Philadelphia,
PA 19122, USA. E-mail: zhan
teng.xie@temple.edu.
Philip Dames, Temple University, Philadelphia,
PA 19122, USA. E-mail:
pdames@temple.edu.
IEEE Robotics & Automation Magazine - December 2022

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2022
