Hi Sarah,
We can't fully answer most of your questions yet since we
are still in the process of collecting requirements for the lab. I would be
interested in knowing what answers to those questions you would like to see. We
want to get feedback from potential users of the lab as we design and built it since those users will
ultimately determine the success of this project.
We have set aside a 12' x 24' area to build the lab. We will instrument
that area with several cameras and a motion capture system. We haven't worked
out specific requirements for bandwidth, latency, etc... We have started
extending the webui package as a potential
user interface but that will take some time to complete. The initial
interface will consist of ssh and the standard ROS packages and will
require endu users to be familiar
with those tools.
The reporting function will be task specific and we haven't finalized
which task(s) we want to do first. I imagine we'll have a separate
program that runs alongside the algorithm being tested and evaluates its
performance. This module could report results through the webui, as log
files or as messages that can be recorded in a bag file. I would also like results to be posted publicly a la
Netflix Challenge. Additional information required for debugging a
specific algorithm would probably be best collected through log files or
filtered bag files.
I
hope this answers some of your questions. As I mentioned above we are very
interested in what requirements users of the remote lab would have. What types
of tasks are you considering for your learning from demonstration research? For
learning from demonstration would you require
remote teleoperation of the robot
base and arms? What additional hardware or software requirements would be needed on the remote
user's end?
Thanks,
Charles DuHadway
Reseearch Engineer
Bosch Research and Technology
Center
Hi Adam,
I know that the Brown group is very excited by this idea. A shared
test bed for domestic service tasks, like those you mentioned, would be an
excellent community resource.
Our group is particularly interested in performing learning from
demonstration experiments in such environments. This would most likely
require a way for users to remotely and interactively demonstrate tasks
directly on the robot.
We are also be interested in how reporting/output would be returned when
algorithms are tested. It seems like rosbags from the PR2 could get
prohibitively large, at least in the current form. Additionally, would the
results from tested algorithms on specified tasks be available to the
community?
This could
potentially address many of the issues the community has faced in terms of
reproducibility. Could you provide more specifics about what Bosch is
planning in terms of space, bandwidth, the user interface, etc.?
Sarah
Osentoski
Postdoctoral Research Associate
Department of Computer Science, Brown University
On Jun 3, 2010, at 2:43 PM, FIXED-TERM Stambler Adam (CR/RTC1.1-NA)
wrote:
Hello All,
As a part of the PR2 Beta Program, Bosch
RTC is setting up a remote laboratory for our PR2. This lab will allow
us to share our PR2 with distant laboratories. This will be similar to
UWA’s Telelab; researchers will be able to log into the robot, upload
code, and perform experiments in a standardized environment in our lab.
Right now, we are in the planning stages for the laboratory and would like
some input from the community. If you don’t have a PR2, what type of
experiments would you like to run on a remote PR2? If you do have a PR2,
are there experiments where you would like to let other researchers test your
algorithm or try out an alternative algorithm in a standardized
environment?
Possible experiments might include opening and
navigation through a standard set of doors, cleaning up a table, and putting
away dishes into a dishwasher.
Any input would be extremely helpful as we plan
and build the infrastructure for the remote lab.
Regards,
Adam Stambler
Robotics Intern Bosch RTC
_______________________________________________
ros-users mailing list