[ros-users] Capability of ARM dev board for ROS host
Austin Hendrix
legotown at aol.com
Fri Jan 13 17:17:41 UTC 2012
My personal robot is running the WG navigation stack with a Hokuyo URG-04LX-UG01 on a pandaboard. I'm running significantly reduced settings for update rates, map size and resolution, local and global planners, and it performs moderately well. I don't have a vision system yet, so I don't know how well it will perform with that additional load.
Prior to the pandaboard, I was using a 500MHz AMD Geode processor with 256MB of ram, and it could almost handle the load.
In both cases I was running a stock Ubuntu kernel without any kind of X interface.
In short, it's possible to make small systems run well, but be prepared to lower your expectations a little and spend a few weeks optimizing the settings for your application.
-Austin
"Armstrong-Crews, Nicholas - 1002 - MITLL " <nickarmstrongcrews at ll.mit.edu> wrote:
>Kent,
>
>Since you're interested in using a beagleboard and/or android phone as a robot brain, I'm guessing you want the robot to be cheap, small, or both. These are actually very different!
>
>
>1) if cheap: spend your time finding/ROS-ifying/writing your own nav and slam packages. I've found move_base and gmapping to be very CPU-intensive (since gmapping is multi-hypothesis, and move_base's local planner is a forward simulator). They work great w/ a consumer laptop (thanks, WG!), but were not really designed for low-spec hardware. Potential field local planner + A* global planner + Karto for SLAM used much less CPU for similar performance.
>
>2) if small: invest in one or more hokuyo laser range finders; or if you really want vision-only, get some panoramic cameras (like PointGrey's ladybug) or a bunch of standard pinhole cameras pointing in all directions (like Google's streetview car), or a more reasonable number of stereo cams (like PointGrey's bumblebee).
>
>3) if both: drastically reduce your expectations of what tasks the robot can perform :)
>
>I would offer a grim warning: beware sending Kinect or uncompressed video or even high-res compressed video over a wireless link (to be processed by a heftier off-board robot brain). Unless you have an uncrowded 300 Mbps connection and are within line-of-sight, your link will get hosed and all tubes will be clogged (not just the /video topic).
>
>Cheers,
>-Nick
>
>PS - @ros-users: is there a better way, e.g. a wiki, to make public these discussions on "how should I make a robot of type X with ROS" ?
>
>From: ros-users-bounces at code.ros.org [mailto:ros-users-bounces at code.ros.org] On Behalf Of Kent Williams
>Sent: Friday, January 13, 2012 4:30 AM
>To: User discussions
>Subject: Re: [ros-users] Capability of ARM dev board for ROS host
>
>Thanks for the thoughts Arjun,
>
>At the moment I have the headless ubuntu 11.10 on the beagleboard so no X window. I'll just end up testing how many substantial nodes I can get running nicely, but ultimately it's looking like these are just not suited to be the main workhorse just yet.
>On Fri, Jan 13, 2012 at 1:03 AM, Arjun Arumbakkam <akarjun at gmail.com<mailto:akarjun at gmail.com>> wrote:
>Hi Kent,
>It depends on what linux distribution you are using on your board. If you use a full fledged Ubuntu distribution, your processor and memory is going to be loaded and there will be little room for other processes.
>That said, if you use a lightweight Ubuntu distribution such as Lxde with Lxdm, you could have room for more computation. We've used a Gumstix Overo board with lxde and lxdm to run a roscore, olsr with two wifi interfaces and a rosnode (with minor computation) and still found the X interface quite responsive. It could have handled some image processing as well, but we ran localization, planning, control and image processing nodes offboard on a beefy laptop like Damon suggested in his email. The Gumstix Overo uses an ARM Cortex A8 core with 512 MB, same as the beagleboard. You could go one up by using just a console image without X. That might give you more room for image processing, etc.
>-Arjun.
>
>
>On Fri, Jan 13, 2012 at 3:38 AM, Kent Williams <k3nt00 at gmail.com<mailto:k3nt00 at gmail.com>> wrote:
>It is our hope to make the robot a standalone unit, and therefore not require any off board infrastructure. I know this is quite a vague question, I'm really just looking to see if anyone has hit the wall big time with the capabilities of these boards or if there is hope for leaning them down enough and slowing down the update frequencies to get by. I have been considering to attempt to use a decent android phone I have as the image sensor and processing resource node using rosjava, since the vision requirements are low. Then possibly the beagleboard or pandaboard would be sufficient for hosting the rest.
>
>
>--
>Kent Williams
>
>k3nt00 at gmail.com<mailto:k3nt00 at gmail.com>
>(818)203-4394<tel:%28818%29203-4394>
>_______________________________________________
>ros-users mailing list
>ros-users at code.ros.org<mailto:ros-users at code.ros.org>
>https://code.ros.org/mailman/listinfo/ros-users
>
>
>_______________________________________________
>ros-users mailing list
>ros-users at code.ros.org<mailto:ros-users at code.ros.org>
>https://code.ros.org/mailman/listinfo/ros-users
>
>
>
>--
>Kent Williams
>
>k3nt00 at gmail.com<mailto:k3nt00 at gmail.com>
>(818)203-4394
>
>_______________________________________________
>ros-users mailing list
>ros-users at code.ros.org
>https://code.ros.org/mailman/listinfo/ros-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ros.org/pipermail/ros-users/attachments/20120113/01cf79e8/attachment-0004.html>
More information about the ros-users
mailing list