|
[DPRG] Re: Allen Robots (Will's emotions)
Subject: [DPRG] Re: Allen Robots (Will's emotions)
From: David M Wilson
davidmw at tx.rr.com
Date: Tue Mar 6 02:02:34 CST 2007
States and behaviors really aren't the goals. These are the poses, stances
and vocabulary much like those taught in karate, dance, and language.
The art and work happen in the transitions between the poses. If the device
actually achieves and holds a 'state' for more than a brief period the
result will seem unnatural.
I'm working to achieve a level of organic behavior animation with a robot
that is designed to be highly interactive and on a conceptual behavioral
model but it isn't ready for daylight. In the current implementation the
focus has been on developing proportional responses to stimuli. Not just
for steering control, but for braking, for excitement, etc.
It isn't clear how well this current implementation will scale and it
doesn't seem to qualify as subsumption. It is fair to say that I'm more
interested in state transitions than achieving classically defined machine
states.
Beyond the 'can you code it' query, one measure of success for behavior
based design is the appearance of organic-like behaviors that are not
intentionally coded.
If achieving specific 'states' or 'behaviors' is your goal then your
definition of a 'flaw' is probably still too rigid for behavior based
designs. 'Programming errors' provide for some of the best organic
behaviors on my bot. For example, the camera aiming code creates annoying
oscillations of the device at certain target distances. Lucky for me, this
PID tuning issue happens to look like the bot is genuinely excited about the
target it just approached. Damp the oscillation and the device loses some
soul. The implementation details really don't matter since the result is
good.
This process could be called programmer assisted learning since there is no
AI. I'm encouraging those behaviors that could develop into a functional or
interactive asset. It is very strange to watch quirks mature into favorite
features. Unfortunately the process still requires a compiler and careful
use of the delete key.
So my bot isn't cheap, isn't an insect, and isn't analog. But it does seem
to qualify as a BBR.
I'll suggest that BBR is more than a software task. If the hardware
platform is imprecise there are more opportunities for unique adaptations
and unexpected results. What is the point in seeking organic behaviors from
a device that utilizes odometrics for navigation? At best you'll get canned
and well rehearsed demonstrations.
Asymmetry, competition for internal resources, conflicting motivators,
spotty navigation skills and physical limitations create challenges that
have solutions we will all recognize.
David Wilson
More information about the DPRG mailing list
|
