If the robot is not sure whether it can complete the
task—for example if the part is "buried" within the bin—it takes
pictures of its situation and calls a remotely located human (the "human
on call") for help.
Like a driver who refuses to ask for
directions when lost, today's industrial robots don't know when they're
in trouble and should stop and get help—which limits their usefulness in
manufacturing.
Now University of Maryland Professor S. K. Gupta and
his students have developed RoboSAM (ROBOtic Smart Assistant for
Manufacturing), an industrial robot smart enough to know when something
is wrong, to pause and to call a human for help.
Currently, industrial robots are used mostly
for high-volume, reliably repetitive tasks with unchanging, tightly
proscribed parameters, such as automobile assembly lines. These robots
are custom-built and programmed specifically for the tasks at hand.
While they excel in such environments, the robots have a limited ability
to assess whether they can successfully complete tasks. The robot
doesn't know it should stop what it's doing if, for example, the parts
it needs are not in the exact position it expects. A chaotic mess can
result—one which humans must then fix.
That's why industrial robots are not used in
factories where high task reliability cannot be ensured. Gupta, a
professor in the A. James Clark School of Engineering, believes a better
economic model would be to give robots the ability to assess whether
they can successfully complete a task, and if they sense they cannot, to
stop and ask a human for help.
His new RoboSAM, based on the Baxter
industrial robot platform, is able to estimate the probability it can
complete a task before beginning it, and can ask a "human on call" for
help if necessary. RoboSAM's abilities may provide a path forward
towards smarter, more versatile industrial robots and more interesting
duties for the humans who work with them.
Gupta's team has successfully demonstrated
RoboSAM in a "bin picking" situation. The robot needs to find a desired
object in a bin of similar objects, pick it up, and deliver it to
another area in a specific placement. If the robot is not sure whether
it can complete the task—for example if the part is "buried" within the
bin—it takes pictures of its situation and calls a remotely located
human (the "human on call") for help. The human then suggests to the
robot what it should do to complete the task, such as stir the contents
of the bin, then try again to locate the needed part.
Gupta believes this work is the beginning of
providing a better economic model for deploying robots, especially for
small- and medium-sized manufacturing companies. "In most situations,
providing task assistance help to robots is much more cost-effective
than recovering from a system shutdown, and it enables humans to move
from doing dull tasks like monitoring and clean up to more challenging
work like helping robots with the tasks with which they struggle."
This research is funded by the National
Institute of Standards and Technology and the National Science
Foundation. Gupta holds a joint appointment in the Department of
Mechanical Engineering and the Institute for Systems Research, and is
the director of the Maryland Robotics Center.
