[ros-users] Hard Realtime requirements and ROS/OROCOS

[Herman Bruyninckx]

On Fri, 14 May 2010, Stefan Kohlbrecher wrote:

> right now, I'm evaluating the ROS-OROCOS integration because my group
> has hard realtime requirements for some control parts of our system.
> The integration works fine so far,  but there are still some things
> that could be easier to do (Automatic timestamp
> conversion/synchronization and generation of OROCOS types from ROS
> msgs for example). From what I gather, on PR2 a RT-PREEMPT Kernel with
> a corresponding SCHED_FIFO Thread is used for realtime control. While
> this seems to work pretty well, one is bound to using RT-preemption
> instead of the dual kernel approach OROCOS permits one to use.
>
> So, without going into much detail, it seems like using ROS only is
> possible, but it might lead to higher latencies in special cases and
> isn't as flexible in the choice of the underlying RTOS. OROCOS on the
> other hand is more flexible in choice of RTOS, but has some more
> overhead (especially in terms of implementation) when integrating it
> with ROS nodes.
>
> Of course, the best of both worlds would be the best thing to have, so
> are there plans for a even tighter integration of both (like the
> aforementioned automatic generation of types)? Does anyone have
> further insights/experience/tips regarding hard realtime using ROS
> with or without OROCOS? Is there some roadmap or cooperation between
> developers regarding this topic?

Yes there is! K.U.Leuven, the main contributor to OROCOS is one of the
recipients in the PR2 Beta programme, with the explicit purpose of
advancing the integration between OROCOS and ROS. So, your
inputs/suggestions/requirements/... come at the right moment :-)
The major responsible person for this integration is Ruben Smits; I suspect
part of the discussions will take place on the OROCOS-dev mailinglist, part
on this ROS-users list...

Herman