|
[DPRG] Robot Code
Subject: [DPRG] Robot Code
From: Randy M. Dumse
rmd at newmicros.com
Date: Fri Mar 16 16:07:41 CDT 2007
Dpa had written:
> I thought a good starting place might be to look at several
> different arbitration schemes, both from published literature
> and practical examples.
I thought you did quite well with this post.
> Note that this technique for arbitration might not work in an
> interrupt driven environment when the motor updates are
> happening asynchronously from the
> arbitrate() loop.
I liked that you showed several ways of doing subsumption, all
still in the basic spirit of the design. I particularly liked
the proviso above about how a simple substitution scheme needs
to be sure the substitutes are substituded before making it to
the outputs, such as might happen if an interrupt routine
asynchronously grabbed the results before every behavior had its
chance to substitute.
> These
> behaviors in a certain sense break the subsumption paradigm,
> because they cannot be subsumed. They are however, in my
> experience, the exception rather than the rule in terms of
> how often they actually control the robot.
>
> Two attributes of ballistic behaviors that may not be
> immediately apparent:
>
> 1. Ballistic behaviors cannot be subsumed: the must run to
> completion.
>
> 2. Ballistic behaviors can , however, be aborted.
However, the above comments on ballistic behaviors breaking the
subsumption paradigm, is purely dpa, and I can find no support
for it in Jones or Brooks. (Which is not to say exactly I
disagree with dpa either!)
In fact, I argue, there is a missing piece in Brooks model, that
a higher behavior must always reset lower ballistic behaviors
before releasing for safety sake, but it becomes a catch 22
issue, because a higher level behavior cannot have a
communications to a lower level that is not an inhibit or
subsume, and that leaves out the much needed reset. It's a hole
in Brooks model. But this is all mine (along with
co-thread-author dan michaels) as we were criticing Brooks on
c.r.m.
So, good job, dpa. Lots of good and carefully stated information
from someone who's blazed that path.
Randy
www.newmicro.com
More information about the DPRG mailing list
|
