On Tuesday 13 November 2007 13:34:14 m [dot] lootsma [..] ... wrote:
> Hello,
>
> I have a question about use of resources.
>
> Does anyone know by experience if it is possible to use Orocos on a system
> such like a humanoid robot with 14 DOF, which need real-time control to
> stay stable, on a 1GHz 256MB PC?

It's safe to say that on such systems, not Orocos (or Linux) will limit you,
but your algorithms will be the main bottleneck. Most RTT and KDL algorithms
are written as such that they don't allocate memory at control-time, such that
they are pretty fast and deterministic, at the expense of requiring more
memory (measured in kbytes, not in Mbytes !).

Peter

On Tue, 13 Nov 2007, m [dot] lootsma [..] ... wrote:

> I have a question about use of resources.
>
> Does anyone know by experience if it is possible to use Orocos on a system such like a humanoid robot with 14 DOF, which need real-time control to stay stable, on a 1GHz 256MB PC?
>
> And to do besides controlling, also need to do a little bit :) of vision, of course with less priority.
>
> Or is it far to less resources? Or does it make a chance.
>
It all depends on:

1. how complex is your control? If you want to run a normal, "ideal"
dynamics control algorithm for a humanoid with 14 DOF, that's not a problem
for such a PC. If you have "nonideal" effects such as friction, backlash,
elasticity, ... your control algorithm could become much more involved.

2. "a little bit of vision": our experience is that vision quickly becomes
a resource killer, unless you do really simple things like tracking a red
ball.

We have no humanoid robot on which we have been able to try out a 14 DOF
control, but we have controlled two 6DOF robots with "a little bit of
vision" in hard realtime on one single PC. (2 GHz, 512MB if I am not
mistaken.)

Herman

Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm