How is the RRP coming along?

8 views
Skip to first unread message

James Ronald

unread,
Jul 31, 2011, 9:32:19 PM7/31/11
to robotrefere...@googlegroups.com
How is the RRP coming along?

I have been playing with ROS on my BeagleBoard currently running
Ubuntu Natty and Linux 3.0. I was hoping to get the TurtleBot robot
package working but ran into some issues in regard to some higher end
ROS dependencies (openni-dev, ps-engine, nite-dev) which will most
likely also need to be built from source. Knowing that HBRobotic has
been meeting at WG, I was wondering if any of the RRP members have my
any progress in this area regarding ROS and the PandaBoard?

Best regards.
James Ronald

Wayne C. Gramlich

unread,
Aug 2, 2011, 12:49:23 PM8/2/11
to robotrefere...@googlegroups.com, Wayne C. Gramlich
James:

I not had a chance to do anything more than download ROS. I have not
event had a chance to walk through the tutorials. I'm not even sure
that I am going to run ROS on my platform, when I stop working on the
mechanical/electrical issues and start programming again.

I'll let other people discuss their progress.

-Wayne

Nathaniel Lewis

unread,
Aug 2, 2011, 11:16:26 PM8/2/11
to robotrefere...@googlegroups.com
Unfortunately, NITE will not work on ARM architecture at the moment.
Despite the open source ideals of ROS, NITE is not open and is a product
of PrimeSense. The Kinect is also based on one of their reference
designs. NITE only available in x86 and x86-64 binaries for Linux, Mac
OSX, and Windows right now. I would personally like to see an open
source project like NITE, but obtaining the quantity of motion capture
data it uses would be difficult. A project would need a decent sum of
donation money and time.

Nathaniel Lewis

James Ronald

unread,
Aug 3, 2011, 10:00:46 PM8/3/11
to robotrefere...@googlegroups.com
Thanks for the reminder regarding NITE. I remember seeing the
mentioned hardware dependencies and was quite disappointed. I have
been thinking about purchasing a dual core 1.8GHz ATOM mini ITX board
just so I can leverage the prebuilt packaged software for my robot.
The price would be similar to a PandaBoard but require about 2 or so
times the power budget. IIRC, the PandaBoard is about 13W and the
Intel D525MW about 30W. It also has a printer port for GPIO ;-)

Or perhaps, my bot doesn't really need NITE. Anyone, know how well
the TurtleBot runs NITE?

Regards
- Jim

Nathaniel Lewis

unread,
Aug 6, 2011, 2:16:29 AM8/6/11
to robotrefere...@googlegroups.com
It has bad performance?  I ran NITE at 30 fps on a Pentium D laptop running at 2.0 GHz.  Its less than twice as powerful as an atom and NITE runs on the Xbox 360 as well.

Nathaniel

Tully Foote

unread,
Aug 16, 2011, 12:45:44 AM8/16/11
to robotrefere...@googlegroups.com
The core ROS packages run fine on the Panda Board with Ubuntu installed.  You just need to follow the compile from source instructions and be patient.  After ROS Electric is released we're going to get back to working on adding arm support to our deb building toolchain. 

I don't know that we've tried running NITE on the TurtleBot.  The viewing point is kind of low to be very effective.  I expect that it would work.    I have heard of several universities working on reproducing NITE's skeleton tracking.  One I have seen discussed at http://answers.ros.org/question/1739/skeleton-tracking-open-source-equivalent-to-closed 

Tully

On Wed, Aug 3, 2011 at 7:00 PM, James Ronald <james....@gmail.com> wrote:

Osman Eralp

unread,
Aug 16, 2011, 1:02:12 AM8/16/11
to robotrefere...@googlegroups.com
How do you find the performance of ROS on the PandaBoard? I ran an OpenCV example (the surveillance demo), and it slowed the board to a crawl. --Osman

Wayne C. Gramlich

unread,
Aug 16, 2011, 5:27:10 PM8/16/11
to robotrefere...@googlegroups.com
All:

My understanding is that OpenCV has some hand coded assembly
code for the Intel inner loops that use the Intel FP accelerator.
The PandaBoard has the ARM Neon floating point acceleration.
To the best of my knowledge, the effort to provide the Neon
FP inner loops has not taken place yet. James Nugen figured
most of the issues out and explained it all to me, but I have
forgotten most of the details.

What needs to happen is that one or two examples of OpenCV inner
loops need to be recoded for ARM, so that the whole process can
be integrated into the ROS build procedures. Once people have
a "supported" example of how to speed up the inner loops, individuals
can code up new inner loops for the routines that they care about.
Note that I have now idea where ARM support is in the ROS priority
list -- I'm just delighted that they are sharing what they have.

Later,

-Wayne

James Ronald

unread,
Aug 21, 2011, 10:42:07 PM8/21/11
to robotrefere...@googlegroups.com
Tully,

That's great news in regard to adding arm support to deb building
toolchain. Let us/me know if there is anything we/I can do to help.
I'm currently running ROS CORE (diamond) and some of the packages from
Brown University on a BeagleBoard xM. I was debating whether to up
grade to a Panda or an ATOM D525 board.

Best regards.
James Ronald

Tully Foote

unread,
Aug 24, 2011, 4:00:24 AM8/24/11
to robotrefere...@googlegroups.com
Hi all,

I've got a bit of an update. We now have the first prototype ROS
continuous integration tests running armel builds of the ROS core
stacks.

http://build.willowgarage.com/view/devel_unstable/job/devel_unstable_ros_lucid_armel/
http://build.willowgarage.com/view/devel_unstable/job/devel_unstable_ros_comm_lucid_armel/

Tully

Reply all
Reply to author
Forward
0 new messages