Rodney Brooks, for those of you who don't know. For those of you who
Cog! His research has been the inspiration for the ROACH project.
Subsumption architecture in it's essence is great, but it is rather lacking
in the ability to deal with complex facts such as relative position, and
higher level reasoning. I propose an addition to the low level subsumption
core similar to Maja Mataric's
work with the 'Nerd Herd'
at Brandeis, combined with a simpler fuzzy-like 'most likely answer' approach.
Low level control is handled by a subsumptive scaling technique, in
which proximity, distance ranging, odometry, compass data, and simple mapping
are combined to form a real-time current position estimation.
Ok, now how do you impliment such a nifty idea on ~16k of RAM (the other
18k is eaten up by IC)? First we can strip away some of the crap, err,
irrelevant code that is built in to IC. One of the problems with using
the 6.270 board is that it was designed for a specific aplication, quite
diferent than a behavior-based programming application.
The inter-process communication available in IC allows for the seperate
subsumption modules to communicate their status to one another.