DPRG List  

[DPRG] Will's emotions

Subject: [DPRG] Will's emotions
From: David P. Anderson dpa at io.isem.smu.edu
Date: Wed Feb 28 13:54:31 CST 2007


So anyway, at last night's RBNO Will suggested the use of the term "emotion"
to describe a level of behavior above the reactive layer of subsumption.  

And used my own robot as an example, to boot!  Ah how sharper than a serpent's
tooth... but I digress.

The idea as I understand it is that "reactive" behaviors are associated closely
in time, i.e., a can touches a bend sensor and the grippers close, an IR detection
on the right slows the wheel on the left, a bumper press reverses the wheels, or
whatever.   The stimulus and response are tightly coupled in time.

The level above the reactive level, which Will would like to call the "emotional"
level, controls the weighting and priority of the reactive behaviors based on
long term accumulation of inputs/states/whatever.  That is, the reaction is
not tightly coupled in time to a particular sensory input.

The example he used was of my SR04 robot going through a sequence of escape
maneuvers trying to get untangled from some knotty situation.  After a certain
amount of time the robot is "aware" that it is not making progress, or getting
too many bumper presses accumlated over time, or whatever, and so it changes
bumper behaviors to a different response (like always turning the same direction
no matter which side the bump, etc) in an effort to escape.

No doubt there are these higher level monitors running "above" the subsumption
layers in all my robots (isn't this Minsky's "meta minds"?) that control and
reconfigure the subsumption layers.  (This allows, for example, the IR avoidance
on SR04, which normally subsumes the sonar, to be subsumed by the sonar when
the robot has observed over time a collection of inputs that "means" it is
approaching a soda can.  Otherwise the IR would never let the robot get close
enough to the can to grab it.)

I think I'd prefer the idea of "harmones" providing large-scale coordination of
simple subsuming layers, rather than "emotions."  Just jargon, really.  But the
word "emotion" has so many other meanings.  In my own thinking I've allways referred
to these routines as "monitors" or the more traditional "watchdogs."

Anybody else coding similar emotional/harmonal layers above the basic subsumption
machine on your robot?  Thoughs?


More information about the DPRG mailing list