uMan video (2MB)
(in the video uMan is still missing wrist and hand)
News:
We have recently brought online a new experimental
platform for mobile manipulation: uMan, the
UMass Amherst
Mobile Manipulor (left, click on the
image for a video of the first moment uMan
became alive; in the video it is still missing its
wrist and hand). It consists of a holonomic XR4000
mobile base (refurbished for torque control) and seven
degree-of-freedom Barrett WAM with a three-fingered
hand. The sensor suite includes fingertip
force/torque sensors, a wrist force/torque sensor, an
arm-mounted camera, a laser range finder on the base
for navigation, and an arm-mounted laser range finder
to support perception using vision for manipulation
tasks.
We believe that uMan is a unique platform that
will enable a new thrust of research in autonomous
mobile manipulation (see this
AMM Workshop).
The construction of uMan was made possible
with NSF funding under contract CNS 0454074.
Mobile manipulation describes the area of robotics
concerned with making robots perform human-level tasks
autonomously, reliably, and safely in everyday
environments. To measure success in mobile
manipulation, we could consider a test, similar to the
Turing
Test, except concerned with work in the physical
sense, i.e, with actually changing the state of the
world. Whereas the original Turing
Test is supposed to assess if computers have
achieved human-level intelligence, our test assesses if
robots have achieved human-level capability of
performing work in their environment; therefore we
call it the Work Turing Test.
A robot passes the Work Turing Test if
it can carry out any set of instructions by
autonomously performing physical work such that the
resulting state of the world is indistinguishable from
the one which would result from a cooperative,
capable, non-expert human carrying out the same
instructions.
Does a robot have to be able to pass the
Turing
Test to pass the Work Turing Test? Probably
not. The Work Turing Test is only concerned with
physical things. I think it is likely that with
proper background knowledge, deduction, and induction
the Work Turing Test is solvable, long before the real
Turing
Test will be passed by a computer.
My research in mobile manipulation is aimed at
creating robots that can pass the Work Turing Test.
Many of the individual research projects presented on
this web site are related to mobile manipulation: motion generation, motion planning, and mobile navigation are all in
service to mobile manipulation. On this page, the
focus is on bringing all of these aspects together to
create an integrated system.
(15MB)
This movie
shows the "mobilization" of a dexterous manipulation
skill. Using this approach, many previously developed
methods for dexterous manipulation can be augmented to
not only be applicable to stationary robotic systems,
but also to mobile manipulators. This is just one
example, but the idea is that much of the research in
robotics has ignored the complexities arising in
mobile manipulation. The video shows how to use
sensory information to address issues of uncertainty,
even if they have not been explicitly accounted for in
the manipulation skill.
