[ros-users] Best option for communication over an unreliable link?

Raymond Sheh rsheh at cse.unsw.edu.au
Mon Feb 14 03:43:38 UTC 2011


  On 13/02/2011 5:15 AM, Ken Conley wrote:
> On Fri, Feb 11, 2011 at 4:40 PM, Raymond Sheh<rsheh at cse.unsw.edu.au>  wrote:
>>   Hi all!
>>
>> We are looking at using ROS on a set of robots, which run autonomously
>> and independently but at times require some collaboration,
>> teleoperation, remote logging and monitoring over a link that is
>> unreliable. I understand that ROS's network communications stack
>> natively assumes reliable, complete connectivity so we'd need to somehow
>> supplement its capabilities.
> roscpp implements a UDP transport as well, so you can do roscpp-roscpp
> unreliable.
>
>> Ideally, as the link degrades we'd like the ability to, depending on the
>> data, selectively drop or buffer and detect (from both ends) what kinds
>> of delays may be present and when the link has dropped completely so
>> that they can react accordingly. We'd also need to have some form of QoS
>> mechanism.
>>
>> Does anyone have any suggestions as to the best way of going about doing
>> this with ROS? If there's already a quasi-standard way of doing this
>> kind of thing within ROS we'd prefer to avoid reinventing the wheel! :-)
>>
>> I'm still getting up to speed with ROS (I've just come from an
>> environment based mainly around Player, Ice and some custom
>> error-tolerant "stuff"). My current thinking is to have each robot run
>> completely separate ROS "clusters" (what's the collective noun for a
>> group of computers/nodes that all talk ROS to each other?) each with an
>> independent master. We'd then develop a node that acts as a
>> communications relay with the ability to provide this kind of graceful
>> degradation and QoS. The "clusters" will talk to each other via these
>> relay nodes. One obvious downside would be that we'd lose the ability to
>> transparently move nodes from one robot to another although given the
>> control we need over what happens on the link I'd assume there isn't
>> really a sensible way of keeping this completely transparent whatever we
>> do.
>>
>> I notice some discussion mid last year about using foreign_relay to do a
>> vaguely similar thing but would I be correct to conclude that the goal
>> of foreign_relay is more to handle the unreliable link in a way that is
>> largely transparent to the nodes involved? Would an alternative be to
>> somehow extend foreign_relay to include the capabilities that we need?
> foreign_relay is a relay node implemented on top of roscpp's UDP
> transport mechanism.  At the very least, it's a good starting point
> for you might relay between different clusters.
>
> As part of our "Building Manager" project [1], we hope to investigate
> other mean of maintaining communication between different robots over
> wifi links, but we haven't started that particular feature
> investigation yet and would appreciate any insight you have.
>
> Hope this helps,
> Ken
>
> [1]: www.ros.org/wiki/Projects/Building Manager

Thanks for your info! I'll have another look at the UDP transport and 
foreign_relay and I'll keep an eye on the building manager project.

I'm still at the stage of getting to know what ROS can do and how it 
fits in with what we'd like to do overall - it'll be another couple of 
months at least before we're in a position to seriously implement 
something.

As for insights, I've got some ideas for what to do (and not to do!) 
based on our previous experience but I am also looking around at how 
other projects have handled unreliable links and QoS issues. I'd 
appreciate any pointers that you (or anyone else for that matter) could 
offer to other projects that are worth looking at for ideas! :-)

Cheers!

- Raymond



More information about the ros-users mailing list