-----Original Message-----
From: [EMAIL PROTECTED] on behalf of Jan Kandziora
Sent: Wed 4/12/2006 6:21 AM
To: owfs-developers@lists.sourceforge.net
Cc: Gregg Levine
Subject: Re: [Owfs-developers] Using One-Wire FS in a robotic system
 
Am Mittwoch, 12. April 2006 05:54 schrieb Gregg Levine:
> Hello!
> Has anyone other then I considered applying the OWFS (CVS or a
> released version) in a robotic system? Comments, criticism, and even
> insults to be delivered off list, and this address.
>
Hi Gregg, why do you want to keep such an interesting discussion off this 
mailing list? owfs-developers is low traffic, and I would find such a 
discussion interesting to the general public here.

Back to the topic:

OWFS has some big pros when it comes to sensors. You could easily place 
hundreds of sensors on the whole robot, making it "feel" the reality very 
accurate. However, as the 1W is "low-speed", one has to think about proper 
preprocessing of sensor data before applying to the 1W. But that may be a 
good idea, too, as the robot has a lot of identical parts that way, making it 
kind of "fail-safe"

On the output side, you will have a bunch of problems with the mechanical and 
electrical characteristics of the actuator itself. To avoid having a 
microcontroller with each actuator, you'll have to use a low-latency host, 
and high-speed host interface, which could do closed-loop steering in <1ms. 
1W and an ordinary Linux host are a mile away from that, especially for more 
than one actuator.

If we talk about self-learning robots, implementing reflex loops with 
microcontrollers could be a goal, too.

So in general, OWFS is good as a "supporting" system to the robot, not as the 
reflex system.

Kind regards

        Jan
-- ----------------------------- --
Nice analysis, Jan. Human sensation is also relatively slow, and works by
summation and rewiring.

To push the analogy further, you could implement "spinal reflexes" -- sensors
feed into regional centers for fine "unconscious control" while a processed
signal is passed upwards.

One interesting aspect would be finding a way to "autoconfigure" or
"autodiscover" the sensors, so that it isn't necessary to hard code each serial
number.

Paul

<<winmail.dat>>

Reply via email to